masahi commented on PR #12949: URL: https://github.com/apache/tvm/pull/12949#issuecomment-1264123972
I agree that (1) is possible, but making such small copies on the device side sounds very slow (e.g. GPU), which would defeat the purpose of removing the host - device round trip. On (2), it sounds like it would add too much complexity to the otherwise very simple code. I'm not sure how such "contiguous region detection" is effective in practice, but if @vinx13 and @junrushao also think it's a good idea, I'm happy to explore this approach. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
